Patient characteristics are described in Table 1. A total of 99 consecutive patients with PCNSL were identified between 2010 and 2021 for whom pre-treatment brain MRI was available for review. Patients were 41% female and 58% male, with median age at diagnosis of 67 years. Average BMI was 29.4 kg/m2, and median ECOG was 1. Patients were treated with a median of 6.5 cycles of HD-MTX and were consolidated as follows: 18 patients were consolidated with myeloablative chemotherapy followed by autologous stem cell transplant, 10 patients received WBRT, 4 patients were consolidated with etoposide and cytarabine , and 67 patients were not consolidated, which includes patients treated with maintenance chemotherapy, targeted agents (ex. Ibrutinib), or immunomodulatory agents (lenalidomide). A total of 52 patients progressed and 42 patients died. Median duration of follow up was 33.3 months. Patients with reduced TMT received a median of 2 doses of HD-MTX throughout the course of their treatment, whereas other patients received a median of 6.5 doses of HD-MTX. With respect to consolidation strategy, only 1/15 patients with reduced TMT was consolidated (with nonmyeloablative cytarabine and etoposide) whereas the remaining 14 were either not consolidated or received maintenance treatments. In contrast, 18/84 patient with normal TMT were consolidated with autologous stem cell transplant, 9 were consolidated with WBRT, 4 received nonmyeloablative cytarabine and etoposide, and 53 received no consolidation or maintenance therapies.
Interobserver agreement for TMT measurements was interrogated using Bland-Altman analysis which revealed a bias of 0.03, calculated as the difference between the measurements from each operator, divided by the average measurement. TMT was normally distributed and fitted to a Gaussian distribution. Mean temporalis thickness was 7.8 mm with standard deviation of 1.8 mm. Patients with TMT less than one standard deviation below the mean (corresponding to <6.0 mm) were grouped as “very thin TMT” for all subsequent analyses. Males had significantly thicker TMT compared to females, average 8.3 mm and 7.5 mm respectively (Fisher’s T-test, two tailed: p=0.044). TMT showed a nonsignificant trend towards inverse correlation with age (r2= 0.036, p=0.059), which was unchanged when including only patients age ≥ 65 (r2 = 0.037, p=0.174).
Currently accepted prognostic scores include the International Extranodal Lymphoma Study Group (IELSG) score  and the Memorial Sloan Kettering Cancer Center (MSKCC) prognostic model . IELSG scores could be calculated for 57/99 patients, due to missing datapoints in the other patients. In this smaller cohort, we were unable to detect a significant difference in survival (Figure 2A; p=0.394) between IELSG 0-1, 2-3, and 4-5.  We also analyzed patients according to the MSKCC score, which divides patients into three groups based on age and KPS. We detected a significant difference in survival between the three groups (Figure 2B, p=0.047), specifically there was decreased survival in patients aged > 50 years and with KPS < 70 (HR = 4.88, 95% CI = 1.84-12.91, p = 0.018), and a nonsignificant trend towards decreased survival in patients aged > 50 years and with KPS ≥ 70 (HR = 2.32, 95% CI = 0.93 – 5.77, p = 0.070). In comparison, when stratifying patients by TMT, we found that thin TMT was highly predictive of both shorter survival (Figure 1B; HR 4.38, 95% CI 2.25-8.53, p < 0.001) and shorter progression-free survival (Figure 1C; HR 4.25, 95% CI 1.95-9.29, p < 0.001). Of note, all 15 patients with very thin TMT had progressed by 13.2 months, and 1 year survival was 5/15 (33%). We subsequently analyzed patients ≥ 65 years (n = 52) and stratified them by TMT. We again found shorter survival in patients with thin TMT (Figure 2A; HR = 2.32, 95% CI = 1.06 – 5.10, p = 0.036) and shorter progression-free survival (Figure 2B; HR = 3.71, 95% CI = 1.72 – 8.01, p < 0.001). These effects were maintained when analyzing only patients <65 years (n = 47), for both survival (Figure 2C; HR 11.1, 95% CI = 3.2 – 38.7, p<0.001) and progression-free survival (Figure 2D; HR = 27.5, 95% CI = 5.9 – 128.0, p<0.001).
Next, we explored the association between BMI and outcomes (Supplementary Figure 1). We hypothesized that elevated BMI would predict shorter survival, and this effect was indeed observed (HR 2.05, 95% CI = 1.10 – 3.82, p=0.028 ). We subsequently hypothesized that the population of patients with high BMI but low TMT would identify a population of patients with baseline obesity and concurrent sarcopenia. We generated an obesity-sarcopenia index calculated as TMT (mm) / BMI *10. The mean and standard deviation were 2.76 and 7.55, respectively. Patients with an obesity-sarcopenia index of less than µ – σ were designated as BMIhigh/TMTlow. There was a nonsignificant trend for shorter survival in BMIhigh/TMTlow patients (HR 1.84, 95% CI = 0.75 – 0.453, p=0.098).
In order to control for confounding by covariates, we performed multivariate analysis including TMT, age, sex, ECOG, BMI, lifetime number of HD-MTX doses, and consolidation as covariates (Table 2). The Cox multivariate model was significant for overall survival (χ2 = 52.12, p<0.001). Of the individual covariates, TMT (HR 4.49, 95% CI = 1.94-10.40, p<0.001), lifetime HD-MTX doses (HR=0.82, 95% CI = 0.74 – 0.91, p<0.01), consolidation (HR=0.15, 95% CI = 0.04 – 0.66, p=0.012) and BMI (HR 1.05, 95% CI = 1.01 – 1.10, p=0.027) were independently correlated with survival (see Table 2). A separate Cox multivariate model was run for progression-free survival (χ2=38.15, p<0.001). Only TMT (HR = 7.87, 95% CI = 3.55 – 17.45, p<0.001) and BMI (HR 1.05, 95% CI = 1.01 – 1.09, p=0.016) were independently associated with progression-free survival (see full results in Table 2).